GW231123: Overlapping Gravitational Wave Signals?

TL;DR

This paper investigates whether GW231123 results from a single high-mass BBH merger or from two overlapping gravitational wave signals. Using Bayesian model selection with four waveform families, the authors compare an isolated-signal model $\mathcal{S}$ to a two-overlapping-signals model $\mathcal{OS}$ and find $\log_{10}\mathcal{B}^{\mathcal{OS}}_{\mathcal{S}}$ values ranging from $\sim 0.21$ to $4.22$, indicating a preference for the overlapping-signal interpretation, though the strength varies by waveform and is tempered by potential noise and waveform systematics. The overlapping-model also reduces discrepancies in recovered source properties between different waveforms, particularly for the louder signal, and highlights the degeneracy with gravitational lensing as a competing explanation. However, a confident claim remains challenging due to non-negligible background from noise and model systematics, and the extremely low prior odds for two simultaneous high-mass BBHs. The work provides a framework to distinguish overlapping signals from lensing and noise, which will become increasingly important as detector sensitivity improves.

Abstract

The recently discovered gravitational wave event GW231123 was interpreted as the merger of two black holes with a total mass of 190-265 $M_\odot$, making it the heaviest such merger detected to date. Whilst much of the post-discovery literature has focused on its astrophysical origins, primary analyses have exhibited considerable discrepancies in the measurement of source properties between waveform models, which cannot reliably be reproduced by simulations. Such discrepancies may arise when an unaccounted overlapping signal is present in the data, or from phenomena that produce similar effects, such as gravitational lensing or overlapping noise artifacts. In this work, we analyse GW231123 using a flexible model that allows for two overlapping signals, and find that it is favoured over the isolated signal model with Bayes factors of $\sim 10^2 - 10^{4}$, depending on the waveform model. These values lie within the top few per cent of the background distribution. Similar effects are not observed in GW190521, another high-mass event. Under the overlapping signals model, discrepancies in the measurement of source properties between waveform models are largely mitigated, and the two recovered sources show similar properties. Additionally, we find that neglecting an additional signal in overlapping-signal data can lead to discrepancies in the estimated source properties resembling those reported in GW231123.

Figures (3)

• Figure 1: Comparison of the estimated source properties when using the isolated signal and the overlapping signals model. We show component masses in the left panel, component spins in top right panel, luminosity distance in the bottom right panel. We label the two signals recovered by the overlapping signals model by their SNRs, i.e., louder (light continuous lines) and fainter (light dashed lines). We compare the XPHM (brown) and the NRSur (green) analyses. When using the isolated signal model, we find discrepancies in the measurements between XPHM and NRSur analyses. These discrepancies are largely mitigated when using the overlapping signals model, e.g., compare the louder signal's source properties between XPHM and NRSur analyses. Similarly, the posterior distributions for the fainter signals agree between the two waveforms, which are significantly broader due to the low SNR of the signal.
• Figure 2: Posterior distributions when we recover two simulated overlapping signals using the isolated signal model with four different waveforms. The top panel shows the measurements of detector frame total mass and mass ratio; with the contours showing 90% confidence interval. The bottom panel show the measurements of luminosity distance. With these simulations, we are able to reproduce the discrepancies in the estimated source properties between different waveforms reported in the GW231123 data release (c.f. Figures 7 and 8 in Ref. LIGOScientific:2025rsn).
• Figure 3: Complementary cumulative distributions (1-CDF) of the background statistic corresponding to the foreground ($\log_{10} \mathcal{B}^{\mathcal{OS}}_{\mathcal{S}}$). The background statistic (black lines) represent how often the isolated BBH signals mimic the features of a pair of overlapping signals. The dashed green (brown) lines show the foreground statistic for NRSur (XPHM) analyses, i.e., $\log_{10} \mathcal{B}^{\mathcal{OS}}_{\mathcal{S}}$ obtained by analysing the GW231123 signal. The background distribution in detector noise is generally above the background of the simulated Gaussian noise, indicating noise artifacts play may mimic the signs of overlapping signals. The distance between the dashed brown and green line may indicate the presence of waveform systematics.